This action might not be possible to undo. Are you sure you want to continue?

# 16

Ulisses T. Mello Paulo R. Cavalcanti1

Mello, U. T., and P. R. Cavalcanti, 2003, A topologically based framework ¨ for 3-D basin modeling, in S. Duppenbecker and R. Marzi, eds., Multidimensional basin modeling, AAPG/Datapages Discovery Series No. 7, p. 255 – 269.

**A Topologically Based Framework for 3-D Basin Modeling
**

IBM T. J. Watson Research Center, Yorktown Heights, New York, U.S.A.

IBM T. J. Watson Research Center, Yorktown Heights, New York, U.S.A.

ABSTRACT

T

hree-dimensional (3-D) basinwide simulation of generation, migration, and accumulation of hydrocarbons has vast potential as an uncertainty- and riskassessment tool in petroleum exploration. To fulfill this potential, several challenges have to be addressed, including the realistic modeling of the evolution of complex geologic structures such as salt diapirs and fault motion. In this chapter, we describe a novel architecture that we have designed and implemented, which specifically addresses technical challenges such as 3-D representation of geologic models, meshing, parallel computing, and visualization of the massive amount of data involved in these simulations. The core of this architecture is a 3-D topological framework for the representation of evolving geologic structures. This enables numeric simulation of geologic processes undergoing large deformations in sedimentary basin and lithosphere. In this framework, the topology (or informally, connectivity) is separated from the geometry of the geologic models, making it possible to update the geometry without altering the model topology. A mesh is treated as a possible realization of the geometric model and hence as an attribute of the topology. This architecture greatly facilitates the automatic meshing and remeshing required for large deformations such as those associated with the formation and evolution of salt diapirs. In addition, this architecture was designed to consider the geometry of geologic elements in the partitioning of the computational domain, and thus, it is suitable to the solution of partial differential equations in parallel. This is beneficial because of the large computational resources required to solve numerically the equations governing heat and fluid-transport processes in sediments.

INTRODUCTION

The simulation of complex geologic processes such as the evolution of sedimentary basins and multiphase fluid flow in sediments is important

1

for the decision-making process in the oil industry. Basin modeling provides a unique framework for (1) knowledge extraction by integrating data from various sources and (2) uncertainty and risk assessment. However, the quality of the assessment is

Also at Federal University of Rio de Janeiro, Rio de Janeiro, RJ, Brazil.

255

Here. or an earth model. Although these systems are appropriate for building static earth models. we describe this framework. The generation of numeric meshes derived from an earth model is a necessary critical step to provide specific data representations such as finite-difference grids or finite-element meshes for the solution of the partial differential equations (PDE) governing heat and fluid transfer in heterogeneous porous media. and seismic ray tracing. In general. they are limited to modeling dynamic geologic structures that evolve through time such as automatic reconstruction of basin evolution. The construction of basin or earth models has been a challenge not only because of the scarcity of subsurface data but also because of the geometric complexity of geologic structures. because some fundamental concepts that appeared in that article are not repeated here for space reasons. fractures. In the following sections. Topological changes occur in few distinct events. Systems such as 3DMove are better fitted for these purposes. The generation and maintenance of numeric meshes can be quite complex for basins that have undergone extensive changes in geometry as a result of compaction. faults. We have designed a framework to build dynamic 3-D earth models for modeling purposes. In addition. 1993) have been designed to build truly 3-D earth models. One important component for achieving accuracy in large-scale basin modeling is the ability to model large deformations such as nonvertical fault displacement and salt motion. . which is used to create a reference model defining a spatial partition for geologic objects. the design of these systems was also oriented toward specific tasks in the industry such as static reservoir characterization.. we focus on the aspects of creating 3-D basin models in our topologically based framework and the management of geologic attributes.’’ Because of the unique nature of geologic data. In this chapter. because they regard earth models as just a collection of 2-D surfaces embedded in 3-D space. has been a major obstacle for the improvement of regional basin modeling studies. The first characteristic is very important. the reference model and the meshes are tightly coupled.g. diapirism.and postprocessing tools to correct inconsistencies to produce a coherent 3-D earth model. The authors recommend the reading of Mello and Henderson (1997). However. which may be severely corrupted with large deformations. we use a topological framework to create and maintain earth models that allow the geometry to be modified during simulations and with new interpretations. These meshes must include the basin geometry and associated physical properties along the evolution of sedimentary basins. we describe a framework that we developed for modeling complex geologic objects with special emphasis on the operators used to build and mesh basin models. salt domes) evolve with time due to various physical processes. because many rock deformations are topologically invariant in time and can be described by geometric changes. and fault motion. the interpreted data are very patchy in nature and provided only a limited amount of information for building the earth model ‘‘puzzle. Multiresolution and multistructure meshes can be associated as attributes to each closed region of the reference model.256 / Mello and Cavalcanti directly dependent on the accuracy of the subsurface earth models and simulations. but they are critical for fully understanding this chapter. During the modeling. Because typical geologic structures (e. Geologic CAD systems such as gOcad and Pyramid (Wiggins et al. Hence.. Note that this framework also provides critical adjacency information for the spatial domain decomposition used in parallel computing. construction of velocity models. The main characteristics of this framework are that the model topology and geometry are kept separated and that the numeric meshes are coupled to the models. as the data permit. The creation and maintenance of evolving earth models and associated meshes. either structured or unstructured. Three-dimensional (3-D) basin modeling requires the best description of subsurface geologic structures. The major sources of data on subsurface geometry come from wells and from interpretation seismic data. including multiresolution meshes. Various commercially available geologic interpretation systems do not enforce model consistency. which are used to generate 3-D raster earth models where a voxel representation is required. it is frequently necessary to use pre. geologic surfaces are frequently inconsistent with each other because of the uncertainties in their generation. commercial CAD systems generally designed for mechanical or architectural applications are not appropriate to build 3-D earth models. and Mello and Henderson (1997) have described techniques to minimize significantly topological changes during basin modeling. making it possible to have a flexible mesh management environment for numeric simulations.

1989). a collection of disconnected voxels commonly represents a fault. Thus. and the main characteristic of this framework is that the model topology and geometry are kept separated. because most of the deformation that occurs during basin evolution is topologically invariant. formity. a mesh is not the model. causing the formation and deformation of geologic boundaries such as uncon- . The topological information is the adjacency description of geologic structures. 1997). these descriptions are not particularly suited to represent geologic objects. to represent Topological Representation of Geologic Objects During the evolution of sedimentary basins. which is used to represent multidomain objects. it does not change the connectivity of the geologic structures. 1994). The first has various geometric and topological operators built in to facilitate the creation of proper 3-D representation of geologic entities such as faults. but just a discrete realization of the earth model. and construction of heterogeneous objects in civil engineering (Cavalcanti et al. or tetrahedral) is just a possible representation of the reference earth model. we designed the framework to create and maintain earth models that allow the geometry to be modified with deformation as long as the topology is kept constant.. Weiler. A complete and general representation of complex geologic structures can be designed using solid modeling concepts (Mello and Henderson. The Modeling Framework Toolkit is composed of two subsystems: Computational Geometry Classes (CGC) and Mesh Generation and Management Classes (MMS). MMS is responsible for generating structured (regular and rectilinear) and nonstructured (tetrahedral) meshes. In the base of the framework. A geologic model can be described by a combination of geometric and topological information. To keep the topology separated from the geometry. it is the connectivity data of the model elements. 1997). and faults. curvilinear. i. It is important to realize that in numeric simulations. 1996) in which each layer has a distinct role in the framework. It also handles registration of meshes generated by other systems. discrete-element modeling. We have designed this framework to specifically build dynamic 3-D earth models. the general structural architecture is modified by various physical processes.. contacts. The geometric information is related to the actual position of model elements such as point coordinates. MMT) to create reference earth models defining a spatial partition.e. and horizons in a 3-D basin model. 1997. This TDS stores all the connectivity information about an earth model necessary to generate and maintain structured and unstructured meshes. salt structures. a user application can use the MMT Application Program Interface (API) to provide access to the model and mesh information simultaneously. the simulation also has access to the topological information of the geologic model. A particular mesh (regular. This kind of representation makes it very difficult to establish the geometry of a fault and the layers that the fault may intersect. Topologically Based Framework The high-level architecture of our modeling framework is shown in Figure 1.A Topologically Based Framework for 3-D Basin Modeling / 257 DESIGN OF THE MODELING FRAMEWORK Although voxel-based geologic descriptions are normally used in the oil industry. This is analogous to the reservoir characterization process in which each reservoir realization is just a possible representation of the reservoir. Similar data structures have been applied successfully in other areas of geologic modeling (Mello and Henderson. We have designed a framework (Modeling Meshing Toolkit. because they cannot handle multiple depth positions for a single x and y position at the surface. Albertin and Wiggins. During the modeling. These changes pose difficulties for defining an abstract representation of geologic structures that is complete and flexible enough to handle substantial changes of sediment geometry and adjacency information during the evolution of sedimentary basins.. making it possible to have a flexible mesh management environment for numeric simulations. because they do not possess any explicit connectivity information on geologic entities. Multiresolution and multistructure meshes can be associated with each spatial domain of the reference model as attributes. we use a topological representation based on the Radial Edge Data Structure (REDS. This feature is critical for basin modeling. Thus. Simple geometric representations are limiting. Informally. crack-propagation modeling (Martha. For example. This is a layered architectural software pattern (Buschmann et al. In the MMT context. we use a topological data structure (TDS) to represent spatial subdivisions defining the geologic objects. 1988). the mesh is the only source of information on geologic models.

It is important to note that meshes are registered as attributes of geologic entities such as blocks. edges. This is essential for increasing the productivity of multidisciplinary projects. REDS is the component that Figure 2. We make extensive use of highlevel topological operators for building earth models. and vertices. Using MMT. and a set of regions may represent geologic layers and fault zones. model. a set of faces or a shell may represent the surface of a fault or seismic horizons. Hence. but only one possible realization of a model or a subregion of the model. stores the topological and geoThe topological entity model is at the highest level of the data structures metric representation of an earth and gives access to the lower-level topological entities such as shells. whereas the Mesh Management and Operators are defined in the MMS. finite volume.. which are composed of an alternating sequence of edge uses and vertex uses (Figure 2). These meshing operators are used to remesh specific regions of the . operators in C++ using an object-oriented design for code reuse. REDS is general and can represent 3-D nonmanifold topology. some degree of remeshing is commonly necessary. This paradigm is very powerful for the simulation of evolving processes where the geometry is changing with time. This is a layered pattern in which the TDS is in its lowest level. In these projects. because TDS are generally too complex to be manipulated directly. the meshing operators can provide multiple mesh representations having multiple resolutions of a given earth model. The associated meshing and remeshing of these geologic objects isbased on the connectivity and spatial subdivision information stored in REDS. We have implemented faces. and therefore. finite differences. whereas the Half-Edge Data Structure used in Mello and Henderson (1997) to represent planar maps can represent only manifold topology in 3-D. The hierarchy of the Radial Edge and Half-Edge Data Structures. The TDS and Topological Operators are implemented in the CGC. or finite elements). REDS explicitly stores the two uses (sides) of a face by two regions that share the same face. this paradigm can also be used to integrate legacy simulation applications. horizons. Edges of REDS may represent well paths. layers. Architecture of the Modeling Meshing Toolkit (MMT). and faults. which controls the generation and management of numerical meshes associated with earth model subregions. The mesh operators and management layers are implemented in MMS. These topological elements are doubly linked the REDS and its topological lists that are also linked to their parent elements.g. a mesh is not the model. Each face use is bounded by one or more loop uses. complex nonmanifold topologies. significant resources are normally necessary to translate the numeric mesh representations among applications that use distinct numeric techniques (e.258 / Mello and Cavalcanti Figure 1. In addition.

Operations between coarse and fine-resolution grids are greatly facilitated in this framework. Each region can have multiple meshes with various resolutions associated with it (Figure 3). and faults represent Figure 3. One important consideration in the MMT design was the ability to perform a geologically based domain decomposition (Figure 3) that can be used for the solution of PDE governing heat and fluid transfer in parallel in a distributed computing environment. It is important to realize that this framework also allows us to manipulate voxel representations (regular meshes) of the earth model with great flexibility. Mesh Generation and Management The structural seismic interpretation of the sedimentary basins provides the geometric elements (set of polygonal surfaces) necessary to create the present-day 3-D basin model and its spatial subdivisions.. where it is commonly necessary to upscale geologic grids to a resolution so that the flow simulation can be executed on available computers (e. create a typical object-based earth model. The resulting earth model contains the space partitions (3-D regions of space) bounded by these surfaces. density.g. Modeling Framework Toolkit concepts. and these meshes are treated as attributes of the model’s region similarly to other physical attributes such as lithology. For example. OPERATORS TO BUILD 3-D EVOLVING EARTH MODELS Both the union and intersection of polygonal surface set defining the boundaries of geologic structures. and velocity. This reference model has spatial subdivisions or regions bounded by the horizons and faults. for example. The faulted block in the upper left represents the geometry and topology of an earth model. Because the earth model has explicit information of the geometry of geologic objects in the model.A Topologically Based Framework for 3-D Basin Modeling / 259 model that have undergone excessive deformation and to transfer information among meshes. This technique has been successfully applied to perform timelapse seismic analysis in the Gulf of Mexico (He et al. we can easily only select. 1997). 1998). . multistructure. or a time line within formations. This kind of decomposition has the potential to provide unique information for load balancing. the seismic voxels of a particular reservoir or source rock object. We are in the process of further enhancing the mesh classes to create and maintain hierarchical multiresolution meshes for upscaling and downscaling operations as well as for solving elliptic PDEs via multigrid methods. we treat 3-D seismic volumes as regular grid attributes of the earth model. A horizon is normally the interface between geologic formations. For each region. given that some problems have more computation restricted to some specific areas of the model such as high-permeability faults or actively deforming areas... The explicit adjacency information that our framework provides can be also used to perform geologically based domain decomposition for parallel computing. as is shown for the region R1 of this model. One particularly important application of these operators is in the area of basin and reservoir characterization. Adding polygonal surfaces sequentially to the model incrementally obtains the geometric and topological description of an earth model. multiresolution meshes may be associated with each. such as horizons and faults. Meshes can be generated for the entire earth model as well as for each individual region of the model. Lagendijk et al.

if the model is not bounded. This is probably the first 3-D present.743 faces. and PLG. In the decimation. model for the Gulf of Mexico Basin that the authors When rectilinear gridded surfaces are provided. and often. a are aware of. Gulf of Mexico Basin model. nonplanar polygons quires approximately 19. gOcad. Geometrically.783 vertices. if topology and geometry. plies these geologic surfaces. name. and they do the topology and geometry of the model. and reframework will work properly. This stems from the . 32. In practice. This model was built from seven input tally.5 MB of memory to store its in a given tolerance (Cavalcanti and Mello. To keep track of the facesurface relationship. age. the result is tolerance dependent. a specified tolerance Processing Operators controls the number and the quality of triangles defining the original gridded surface. The most complete format is discontinuities caused by the sliding of fractured virtually a map of the model internal representation rock. mapping directly positioning of surfaces in the 3-D space. 1990). A face may be present in multiple-input surfaces (if the defining polygons have a tangent overlap). insertion of a bonding box. it 32. This model possesses 40. they do not dedefines the spatial partition of the earth model and fine closed geologic objects. the order of inserting faces is irrelevant and the result is deterministic. To improve the execution time for the spatial localization of the face set in the model that can potentially intersect the incoming polygon. (1997).743 faces and Following the model creation. 40. we use a dynamic index structure given by an R*-tree (Beckman et al.783 vertices. are automatically triangulated. and we have built. Unfortunately. each face has an attribute index that points to its respective entry in the model attribute table. input surfaces are frequently The process of incrementally adding polygons incomplete and noisy. This model was built from seven closed region of interest encominput surfaces and generated six major regions in the space after the passing all the surfaces. To ensurfaces described in Mello and Karner (1996). a bounding box is added to delimit a Figure 4. which contains all the surface attributes such as type. The frameintersect each other. and it sure that the planar intersection engine of the has six regions. 1999). decimation filter is normally applied to compress and reduce the redundancy in the object geometric information. After all the surfaces are inputted. as described in Cavalcanti et al. In principle. A set of planar polygons or regwork can also export other popular 3-D formats inular 2-D grids embedded in a 3-D space usually supcluding VRML.. there are no restrictions on the in the computer memory to the disk. and it represents the entire Gulf of each of its defining polygons is inserted incremenMexico Basin. Figure 4 is an example of one of the earth models The inputted polygonal surfaces are read. Note the complexity of this realistic model at the can be saved in several formats border of the basin. etc.260 / Mello and Cavalcanti generates the set of faces bounding a geologic object. with various degrees of trade-offs between file size and explicit connectivity information.

and closure problems are associated with mismatches between surface boundaries An interface is a set of faces adjacent to two dissuch as gaps (where a surface is too short) or rims tinct regions. regions of the earth model is useful in the parallelGeometric ‘‘noise’’ can induce the creation of subization algorithms that use unstructured meshes. . (b) Result. where the interface is simply removed. we briefly describe the most important high-level opThe merge region operation removes the intererators. dramatically improve the productivity when creatMerge Region Operator ing high-quality basin models. we find its lamina edges. Regularize Region Operator Regularization is the process of deleting all dangling faces in selected regions of the earth model.A Topologically Based Framework for 3-D Basin Modeling / 261 actively. (a) A simple tetrahedral region generated by combining four ing it. as have designed several high-level operators that can described next. It is normally used to remove excess rims that were generated by the intersection of mismatched surFigure 5. or shell. This operation top of the low-level topological (Euler) operators deis necessary to join region parts of a geologic object scribed by Weiler (1988) and summarized in Mello that is supposed to be continuous. lamina edges bound dangling faces. based on the fact that a dangling face in the rim is adjacent to just one region. In the next items. A series of closed cycles of lamina edges defines the boundaries of holes and the external boundary of a surface. The holes occur beExtract Interface Operator cause of lack of data. as opposed to faces on the interface beuncertainty in the interpretation process. (b) The resulting enclosed region after the be paved is normally done inter. and hence. this overwhelm the model with unusable information. This operation is paved. we the model. because they through their common interface.regularization operation. The selection of the holes to rectangular polygons. Inaccurate geomand Henderson (1997). In addition. The most tween two regions. Pave hole operator application. (a) A hole in the surface is faces (Figure 6). clos. common problems are holes in the surface and a lack of closure among surfaces. The ability to extract adjacency between (where a surface is too long). operation is necessary to merge adjacent regions of To address the issues related to geologic data. enforcing the closure of holes only in selected places. A lamina edge is an edge that has just one face incident to it. The pave hole operator lists the lamina edge cycles and allows a face to be glued onto one of them. The selection of the Pave Hole Operator regions to be merged can be done interactively or To close holes in a given surface. bestantial number of very small regions. etry of the input surfaces can unwillingly create discontinuous geologic objects. Figure 5 displays a typical result of this operation. These spurious cause the interaction between the two regions occurs small regions should be eliminated. of the model. Notice that they are implemented on the face between two adjacent regions.Figure 6.

To extract a cross section. triangular. rectilinear. we generated with 3034 regions because of geometric sample the geologic objects by shooting rays at specnoise. (b) of the 3-D geologic objects on the Triangulated cross section of the Gulf of Mexico model (Figure 4). These operators are used for numeric solution of PDE and for interpolaFigure 7. Then we use the intersection engine of our framework to define the Figure 8. Foley et al. The previously generated automatically using the adjacency and attribute lower-dimensional meshes constrain the higherinformation. spurious small regions in the model. Regions with a volume less than a fraction of the model volume can Regular/Rectilinear Mesh Operator be automatically aggregated to produce larger adjacent regions with similar attributes (Figure 7). . A small region (red) (a) before and (b) after its aggregation to tion purposes. moving to higher dimensions (3-D) hierarchically. The This operation is extensively used where the surunstructured meshes are undoubtedly the most imfaces have geometric noise that creates numerous portant in representing complex geologic structures. because they are not part of any of the geologic objects. start on cells of lower dimension (1-D). dimensional ones. The small region aggregation operator that aggregates reare hierarchical in nature. samples representing grid cells are generated sections of an earth model is a very useful operation.5-D Cross Sections tween a ray and the model surface faces are sorted along the ray path in one of the coordinate directions. Meshing operators a salt dome. In the We take full advantage of the explicit adjacency process of aggregation. all the spurious small ified intervals that define a rectilinear mesh. Given an arbitrary meshThe Gulf of Mexico model (Figure 4) was initially bounding box in the space of the earth model. a plane that cuts the model must be supplied. Applying this operator. (a) 2.. because several applications used in the industry still need 2-D models. This regions were eliminated. in computer graphics for visible-surface determination (e.. The intersection points beExtracting 2-D and 2. Meshing Operators Meshing operators are used to create alternative discrete realizations of the model or regions. and gions with a volume less than a specified threshold performed this most of the meshing operations aggregation.5-D model is a 2-D model with constant thickness in the third dimension. regular meshes efficiently. These faces are not inserted in the earth model.g. The most important meshing opSmall Region Aggregation Operator erators we implemented are regular. as expected initially. 1995). the extraction of 2-D planar cross Then. tetrahedral.262 / Mello and Cavalcanti cutting plane (Figure 8). and curvilinear (partially).5-D regularly gridded cross section of a salt dome. Resulting sections from the application of the cross section faces that represent a 2-D map operator. A 2. with only the relevant six process is analogous to the scan-line algorithm used regions remaining. and it is useful to test new 3-D algorithms comparing it to 2-D solutions. the interface between the reinformation stored in our framework to generate gion to be removed and an adjacent region is deleted. either for visualization purposes or for simulation. For modelers.

Tetrahedralization Operator This operator uses the 3-D Delaunay approach to generate constrained tetrahedral meshes (Figure 9). which used generating initial triangulated surfaces either (1) for the automatic point creation operator to generate further refinement in cross sections (Figure 8) or (2) points in the salt domain.000 elements generated for an faces of the model. method of generating points in the domain. Automatic Point Creation Operators Although Delaunay triangulation has a well-defined method Figure 9. This operator is critical for basin modeling in our framework. (a) MMT generates a regular mesh with the same resolution of the seismic volume. In the present implementation. as described in Cavalcanti and Mello (1999). as described by Shewpoint creation. we show an example in which we extract the seismic voxels of Figure 10. (b) Tetrahedral to create the connectivity of a set of points.A Topologically Based Framework for 3-D Basin Modeling / 263 three reservoir objects. a 2-D grid cross section generated by domain. (b) Mask operations are performed to extract only the reservoir attributes. because tetrahedral meshes are used for the solution of PDE as well as for interpolation. We triangulated this model. cubic and hexagonal lattices to create the background Triangulation Operator meshes (Mello and Cavalcanti. we used this operator is displayed. 1991). These operators are Laplacian smoothing and Smart smoothing. and representation of geologic structures with complex geometry. the R*-tree is used the quality of tetrahedral and triangular meshes. In Figure 10. Figure 11 shows a 3-D tetrahedral mesh with This operator triangulates any set of polygonal approximately 40. In Figure 8. Because of the difficulty in for creating triangulated bounding surfaces that will visualizing 3-D meshes. chuk (1997) and by Joe (1989. To triangulate a set of faces. it does not provide a mesh of the same layers. . we used Delausection using the cubic lattice as a template for the nay triangulation techniques. we use Figure 12 to show a constrain the tetrahedralization of regions in the cross section of this model. discussed Mesh Extraction and Merge Operators These operators manipulate voxel representations of an earth model using explicit information of the geometry of the geologic objects in the model. Mesh Smoothing Operators We have implemented operators for improving the quality of tetrahedral meshes. We to locate quickly the model faces along the shooting used background meshes to create the points in the ray. For efficiency. (a) Reference model for a set of stacked layers. The attributes of these voxels can be used for the process of reservoir characterization and similarly in the process of source/reservoir rock characterization in basin models. The creation of these points in the domain may be necessary to improve along the ray path. This operator is very useful in earth model composed of six regions. Extraction of 3-D seismic attributes from three reservoirs defined by a reference model. isosurface generation. 2000).

In Figure 14. (b) Mesh see that the peak of the minimum generated using cubic lattice as the background mesh. we display the mesh of the model shown in Figure 12 after the applications of Laplacian smoothing and Smart smoothing. Smart smoothing (dotted line) is not that effective for 2-D triangulations. one can Figure 12. Based on this typical example. Smart smoothing the salt dome region with approximately 40. (a) Input surfaces for a model and (b) a tetrahedral mesh for invalid meshes. the new location of each node is determined by averaging the coordinates of the nodes connected to it. significantly shifting the peak of the minimum angle to approximately 528. In 3-D. This algorithm performs very well for 2-D triangulations. it is clear that the background mesh has an import impact on the final quality of the smoothed mesh. in detail by Freitag (1997). angle distribution is close to 438 (solid line).264 / Mello and Cavalcanti constraining boundaries). These operators are very useful for improving the quality of elements that have been severely distorted during deformation. but in 3-D. Hexagonal lattices have perfect equilateral triangles. In Laplacian smoothing. In Figure 15. we show a mesh for the cross section using a Figure 13. (a) Cross section of the model shown in Figure 11. it may produce Figure 11. Laplacian smoothing (dashed line) improves this mesh. Smoothing techniques In Figure 16. a ing the nonfixed nodes (which are not part of the . The effect of these operators can be seen in Figure 14.000 tetrahedra. respectively. In Figure 13. fixes this problem by making motions that only increase the minimal dihedral angle of the set of tetrahedra connected to the node to be moved and produces valid meshes. The triangles generated after the Delaunay triangulation of a predominantly cubic lattice should have minimal angles of 458. but it produces triangles with a better size distribution. one can see how much the hexagonal reduces the tetrahedral element distortions by movlattice is superior to the cubic counterpart. all with internal angles of 608. Cross section meshes after the applications of the (a) hexagonal lattice as the template Laplacian smoothing operator and (b) Smart smoothing operator. for the automatic point creation.

In addition. . e. 2002). which is the partition of the space in which each partition cell represents the space close to a vertex of the mesh (Figure 18). as described by Smith et al. Mesh Dual Operation (Voronoi Tessellation) Delaunay meshes have the property that the circumsphere of each tetrahedron contains no other vertex other than the four vertices of the tetrahedron. This operator uses the information from the Delaunay meshes to construct Voronoi tessellations. Mesh generated using a hexagonal lattice (a) before and (b) after smoothing. Schlogel et al.. search.g. edges.A Topologically Based Framework for 3-D Basin Modeling / 265 Figure 14. Graphs can also represent meshes (e. hexagonal lattice does not produce perfect tetrahedra with all internal dihedral angles of 728. not only for the factorization of sparse linear systems but also for parallel computing algorithms such as iterative substructuring. it is necessary to create meshes that follow the stratigraphy of geologic layers. Figure 15. Histogram of the distribution of the triangle minimal angle in meshes before and after applying smoothing operators. Histogram of meshes shown in Figure 15. These methods are particularly attractive. with pointers linking together an object’s faces. in which faces and vertices occupy complementary locations and the position of each dual vertex is computed as the center of mass (centroid) of the mesh cells. these algorithms can be used to combine model regions for load balance in parallel computing. it is simple to construct adjacency matrices for any topological entity of the model by querying the REDS. and vertices.. Node Ordering Operators The spatial partitions represented by the REDS TDS. which renumbers the nodes hierarchically from region internal nodes to the region interfaces. (1996). This operator creates unstructured meshes that respect time lines. in our framework. can be transformed into an irregular graph-based model.g. as depicted in Figure 19. Figure 16. The Delaunay triangulation is the dual of the Voronoi tessellation.. A dual of a mesh is commonly defined as another mesh with the same topology (genus) but different connectivity (vertex-face incidence). we are using a given model spatial partition as a geologically based domain decomposition for substructuring methods. and finitevolume simulations. Just as for graphs. Stratigraphic Mesh Operator In many circumstances. Figure 17 shows a typical sparse matrix resulting from the application of a node-reordering operator. because graph algorithms can be used to reorder mesh nodes for good orderings of sparse matrices arising from the discretization of PDE. These tessellations are used extensively in geostatistical declustering algorithms. This is important.

. This technique is general. Each topological entity points to a geologic attribute index of the table. (a) Convex hull of a 3-D model and (b) its corresponding mesh dual. or Voronoi tessellation. To generate this figure. as described in Mello and Henderson (1997). in combination with activation and deactivation of faults. Layer Operator This operator traverses the regions of the model and returns a set of all the regions that are between two specific horizons. can be applied to regions with complex geometry. as depicted in Figure 3. it is more natural during the modeling to use geologic entities such as layer and blocks rather than topological entities. The structural block classification is described in Mello and Henderson (1997).266 / Mello and Cavalcanti A set of operators has been designed to traverse the TDS and create sets of topological entities representing the corresponding geologic objects. and difference. intersection. The use of the present-day basin topology in combination with ‘‘activation’’ and ‘‘deactivation’’ of faults present in the model greatly reduces the topological Geologic Operators Although the TDS provides a solid representation of geologic models. This set represents a geologic layer. Fault Tree Operators These operators create and maintain a general tree of faults in the model. and is especially useful for creating elements to simulate sediment deposition. we have applied the operator to create a statigraphic mesh following the time lines of this particular sigmoid. The union of blocks is necessary where blocks are moved kinematically. Figure 17. The mapping between topological and geologic attributes is achieved by keeping a table of the geologic attributes in the model. The operator is used for visualization as well as to strip out or to activate a layer in the model. Note that the region internal nodes are concentrated in the upper left side. The isochron surfaces are then used to constrain the generation of a new mesh that will contain these constraining surfaces. Node connectivity matrix resulting from the application of the node reordering operator using the adjacency information from the TDS. which are bounded by the model-bounding box or by faults. Figure 18. Constructive Nonregularized Geometry Operators During the modeling. This set of regions represents a block. This is done by first generating a background mesh that is used to solve the Laplacian equation to find isochron from the geologic ages defined as boundary conditions of a region. These blocks can be manipulated using normal set operations such as union. it is important to minimize the topological changes in the basin model. Block Operator This operator returns a set of regions stacked vertically. The fault tree is used to classify geologic blocks with respect to the faults controlling the block motion during the modeling process.

The pressure field and associated streak lines of fluid flow surrounding the salt motion was performed kinedome. These operators are described in detail in Cavalcanti et al. operators are used to update the topology of the . In this example.A Topologically Based Framework for 3-D Basin Modeling / 267 Figure 19. changes in topology that occur in the mesh. the tetrahedral mesh representing the salt changes during the modeling. This geologic scenario was inspired by the salt tongues present in the Gulf Coast (e. When this curs. as shown in the previous mesh were interpolated over the new Figure 20. The salt dome evolution starts from a flat surface corresponding to the top of the salt that evolves through Figure 21. b) Two stages of a salt dome (magenta) evolution. while conserving the total salt mass. This is a and (b).. The model geometry before and after the collision (piercement in this case) is unified. the change in the topology is calculated automatically using synthetic example to demonstrate constructive nonregularized geometry operators. the surfaces used to define the final topology of the model are shown in Figure 11. In our framework. Here. Mello et al. Unstructured stratigraphic meshes on the right can be generated by constraining the mesh generated with isosurfaces representing time lines. When the ing an evolving salt dome using geometry of the geology is defined by two steps of a reconstruction. topological changes may be required. However. This process was automatic. one application of our modeling framework is shown. AN APPLICATION: HEAT AND FLUID TRANSFER SURROUNDING AN EVOLVING SALT DOME In Figure 21. without any user model. we calculate the temperature and pressure field surroundFigure 20. Any internal boundary in a continuous geologic entity generated in the unification is removed. (a) finite-element methods. Constructive nonregularized geometry (CNRG) intervention. These time lines on the left are obtained by solving the Laplace equations with geologic ages of the horizons as boundary conditions.g. the capabilities of this framework for modeling complex structures such as a salt dome with overhangs. For examoccurred. As the salt deformed. (1997). and new regions generated during the unification are classified according to the geologic attributes. where salt had to be remeshed numerous times because of exmotion is modeled or where substantial erosion occessive deformation of some elements. matically. these operators are used when a change of topology is detected by collision. the pressure and temperature fields from ple. model automatically. (a. 1995). (c) the time. as shown in Figure 21. Two stages of a salt dome piercing a geologic layer. in the process of salt piercement..

U. p. Addison-Wesley.-P. 4-D seismic simulation of a complex turbidite sand. REFERENCES CITED Albertin. Freitag. Joe. Anderson. These contributions were important in the early stages of designing our modeling framework. N.268 / Mello and Cavalcanti Note the deflection of the fluid path around the low-permeability salt. Three-dimensional constrained Delaunay triangulation: A minimalist approach. p. F. 59 – 63. thereby moving the fluid away from reaching the reservoir (in green). October 10 – 13. P. Lagendijk. September. p. AMD. A. 10. 37 – 43. S. L. and simulate and visualize evolving 3-D geologic models. P. G. and ENI-AGIP. K. CONCLUSIONS In this chapter. 987 – 997. Sommerlad. which enables numeric simulation of geologic processes undergoing large deformations in sedimentary basin and lithosphere. Rohnert. 28. Mello. 31. 1994. Tom Jackman. Embedding geologic horizon surfaces in tetrahedral meshes for geologic modeling: Society of Exploration Geophysicists 64th Annual International Meeting.. 220.. 209 – 220. T. The framework incorporates a series of techniques that represent meshes. P. p. In our opinion. R. Structural reconstructions provide insights to the dynamics of the basin. Upscaling: A tool from the past?: IBM Research Report RC20985. 1999. p. 1997. v. we thank Wendell Wiggins for the comments and insights related to constructions of 3-D earth models. 1175 p. which partially funded ´ this research work. and U. Killough. and U. Computer graphics: Principles and practice. In this framework. the most challenging problem is the utilization of data from 3-D kinematic reconstructions in dynamic simulations. Foley. R. N.: New York. and B. In addition. John Wiley & Sons.. Buschmann. 1990. this framework was designed to consider the geometry of geologic elements in the partitioning of the computational domain. 1995. H. Stal. Numeric meshes are treated as a possible realization of the geometric model and hence as an attribute of the topology. Delaunay versus max-min solid angle triangulations for three-dimensional mesh generation: International Journal of Numerical Method in Engineering. and Kate Chess. in Proceedings of the 1990 ACM SIGMOD Conference on Management of Data. p. World Oil.. Wiggins. B. van Dam. Boulanger. Beckman. the topology is separated from the geometry of the geologic models. C. P. Applied Mechanics Division. R. 1997. B.. Mello. and J. p. Hughes.. and W. R*-tree. May 23 – 26. 1994. R. Cavalcanti.. 1998. Mello. Guerin. and thus it is suitable for the solution of the PDE in parallel. E. J. He. Basic to this framework is a 3-D topological representation of evolving geologic structures. 1997. we have described a modeling framework that addresses several challenges in modeling the evolution of complex geologic structures such as salt diapirs and fault motion. R. 1991. K. and Andy Conn for critically reviewing this manuscript. T. Schneider. dynamic simulation of salt structures rarely succeeds in reproducing in detail the salt shapes observed in the field. 718 – 741. A. (1995). Cavalcanti. v. We intend to use a hybrid approach combined with domain decomposition to address this problem in the future. 1990. F. This framework greatly facilitates the automatic meshing and remeshing occurring during large deformations such as those associated with the formation and evolution of salt diapirs. Seeger. New Jersey. 497 p. 1996. but the process of backstripping is subject to various problems because of the difficulty in correcting for dynamic processes such as overpressure development. 4. 322 – 331. Expanded Abstracts. H. There are still remaining challenges that have to be addressed to perform dynamic simulation of sedimentary basins. and L. v. However.. W. 502 – 505. Feiner.. v. Martha for providing an initial implementation of REDS in C and Claudio Esper´ anca for making available his implementation of the ¸ . in Trends in unstructured mesh generation: American Society of Mechanical Engineers. and M. P.. Los Angeles. 1999. D. 1989. making it possible to update the geometry without altering the model topology. Three-dimensional triangulations from local transformation: SIAM Journal on Science and Statistical Computing. ACKNOWLEDGMENTS We thank Luis F. Non-manifold modelling: An approach based on spatial subdivision: Computer-Aided Design. Lake Tahoe. California. Carvalho. Petrobras. Meunier. in 8th International Meshing Roundtable ’99. California. The R*-tree: An efficient and robust access method for points and rectangles. Pattern-oriented software architecture: A system of patterns: New York. and U. October 23 – 28. F. Martha. J. 2nd ed. T. Atlantic City. 16 p. On combining Laplacian and optimization-based mesh smoothing techniques. A. Kriegel. no. This behavior of the heat and fluid flow is expected and has been described before by Mello et al.. We are very grateful to Western Atlas International. Joe. In particular.

1989. 1997. 2000.D. L. p. 2002. J. Fox. A. D.. B. K. New York. 1993. J. California. and M. p. and P. 1988. Pennsylvania. and J. G. Mello. W. 170 – 174. thesis. Karypis. Domain decomposition: Cambridge University Press. E. 343 p. D. Foster. 14. U. Shewchuk. R. no. Karner. Cornell University. P. 1996.. 697 – 716.A Topologically Based Framework for 3-D Basin Modeling / 269 Martha. eds. September 26 – 30. in M. K. in J. Dongarra. Kumar. 12. U. The role of salt in restraining the maturation of subsalt sediments: Marine and Petroleum Geology. Sourcebook of parallel computing: San Francisco. and W. Karner. Ithaca. Topological and geometrical approach to numerical discretization and arbitrary fracture simulation in three-dimensions: Ph. October 2000. Graph partition for scientific simulations. U. and W..D. T. 1367 – 1396. G.. Development of sediment overpressure and its effect on thermal maturation: Application to the Gulf of Mexico Basin: AAPG Bulletin.. 253 – 261. E. and G.C. Schlogel. Expanded Abstracts. v. Stankovic. Mello. A point creation strategy for mesh generation using crystal lattices as templates: Proceedings of the 9th International Meshing Roundtable. The radial edge structure: A topological representation for non-manifold geometric boundary modeling. Pittsburgh. Gropp. Mello.. Delaunay refinement mesh generation: Ph. Bjorstad. 1995. 207 p. Mello. K. 80. L. Albertin. U. Geometric modeling for CAD ¸˜ applications: Amsterdam... 3 – 36. T. and V. T. Elsevier Science Publishers.. N.. 1993. L... J. I. p. Techniques for including large deformation associated with salt and non-vertical fault motion in basin modeling: Marine and Petroleum Geology. v. D. Carnegie Mellon University. 1996. Gropp. J. p. Kennedy. 1997. . Morgan Kaufmann. Cavalcanti. 551 – 564. F. and G. eds. U. G. Washington. 224 p. Lousiana. Wiggins. Smith. Weiler. Anderson. Torczon. T. v. White. Building 3-D depth-migration velocity models with topological objects: Society of Exploration Geophysicists 63rd Annual International Meeting. thesis. McLaughlin. Henderson. D.. Encarnacao. F. p. L. p. 5. New Orleans. R. Wozny. and R.